Nvidia Vera Rubin NVL72 measured to have 10x TPS per megawatt than Blackwell (twitter.com)

CoreWeave has unveiled the first measured performance results for NVIDIA's Vera Rubin NVL72, showcasing a remarkable improvement of 10 times the tokens processed per second (TPS) per megawatt compared to the previous Blackwell architecture. This breakthrough was achieved using the DeepSeek-R1, marking a significant milestone for NVIDIA's capabilities in AI processing. The results are not just projections; they represent real, live hardware metrics that underscore the advancements in energy efficiency and computational prowess of the Vera Rubin architecture. This improvement is significant for the AI/ML community as it addresses one of the pressing challenges in modern AI training—energy consumption. Boosting performance while reducing energy requirements could pave the way for more sustainable and efficient AI applications, making high-performance computing more accessible and affordable. The specific increase in TPS per megawatt indicates that the Vera Rubin NVL72 is poised to accelerate AI workloads dramatically, enhancing the ability to process vast datasets and complex models, which is crucial for advancements in various AI fields.